Modeling changes in assessments to predict needs and guide care planning in a home care setting

Transcription

1 Modeling changes in assessments to predict needs and guide care planning in a home care setting TVN Mindmerge Victoria Jeff Poss Debra Sheets Stuart MacDonald Cheryl Beach

2 Purpose of Study Develop a predictive frailty measure for seniors receiving home care by applying advanced analytic methods to data from the RAI-HC Specifically: 1) using available data, develop predictive models of frailty that can inform decision-making 2) implement the frailty measure with healthcare professionals working with home care clients 3) demonstrate the impact of the frailty intervention on decision-making, services utilization and health outcomes

3

4 Resident Assessment Instrument (RAI) Overview Standardized, comprehensive assessment and care planning system Minimum Data Set (MDS) Origins: US Nursing Homes Medicare/Medicaid (1987 Omnibus Budget Reconciliation Act) Public domain in U.S. Elsewhere copyright held by Versions for other settings later developed: home care, inpatient psychiatry, acute care, post-acute care, community mental health, palliative care, intellectual disability, etc.

5 Example items (from RAI-Home Care) ~300 items - Trained assessors - Use all available sources of information - Client - Informal caregivers - Referral source, medical record, etc.

6 RAI-Home Care Long-stay (~60 days or longer), adult, non-palliative At intake, then re-assessed every 6-12 months or sooner if significant change Individual level: Care planning Track outcomes over time Aggregate level: Understanding served populations Outcomes related to service quality Expected resource intensity (case mix) Program evaluation/research

7 RAI-HC in Canada National reporting standard, adopted by the Canadian Institute for Health Information (CIHI) in 2001, part of the Home Care Reporting System CIHI provides training, reporting, data standards, and a national data repository Mandated in 8 provinces/territories: BC, AB, SK, MB (WRHA), ON, NS, NL, YT At least 3 million assessments done in Canada to date ~400,000 assessments/year

8 RAI-HC informed snapshot (HCRS Quickstats, CIHI) 2014/15 BC ON NL YT Average age Female 64% 64% 64% 63% Alzheimer s/dementia 36% 25% 16% 14% Have informal caregiver 96% 97% 90% 92% Informal care hours/week Receive extensive help with ADL Mild or greater cognitive impairment 19% 20% 12% 3% 61% 58% 27% 37% Signs of depression 21% 23% 17% 26% Antipsychotic medication 11% 15% 8% 7% CHESS* 2 or greater 32% 46% 27% 29% * Changes in Health, End-stage, Signs and Symptoms: marker for health instability

9 Conceptual Overview RAI-HC data for tracking change in status over time, but the predictive value has been largely untapped RAI-HC outputs have generally been limited to evaluating outcomes for a single point in time We will mathematically model an individual s change in RAI-HC measures to predict likely changes in the level of care required for an individual

10 Performance Clinical Decision Making: The Central Importance of Change Decision threshold Decision point Red: Precipitous decline Time

11 Performance Clinical Decision Making: The Central Importance of Change Green: Stable, but variable Decision threshold Decision point Time

12 Performance Clinical Decision Making: The Central Importance of Change Decision threshold Blue: Gradual decline Decision point Time

13 Performance Clinical Decision Making: The Central Importance of Change Decision threshold Decision point Green: Stable, but variable Decision point fails to use prior assessment information could it be helpful in predicting future performance? Blue: Gradual decline Red: Precipitous decline Time

14 Present Study: Key Research Objectives Model individual differences (intercepts) and change (slopes) for select subscales and indicators of the RAI-HC computed scales clinically-informed combinations of RAI-HC items (e.g., living alone + cognitive impairment) individual RAI-HC items (e.g., stamina) These intercepts and slopes, along with other covariates (e.g., age, caregiver status), will help predict risk of transition to subsequent levels of care (e.g., HC to assisted living)

15 Present Study: Quantitative Models Statistical models that facilitate a joint estimation of survival (e.g., time to transition or death) and change (growth curves) components of the longitudinal data These shared growth-survival parameter models permit the simultaneous estimation of intercepts/slopes of RAI indicators predict clinical outcomes McArdle et al., 2005

16 Project Plan Obtain de-identified data Island Health (exploratory) Canada (CIHI) Modeling Quantitative & clinical expert guidance Test predictive indicator with selected home care case managers and ~30 clients over 3 months, within Island Health

17 Acknowledgments This research was funded by a grant from the Technology Evaluation in the Elderly Network

18 Questions Jeff Poss, Ph.D. University of Waterloo (Adjunct)